Mission
FamilyOS is building the world’s first AI Operating System for Families.
Our mission is to help every family make more confident decisions together, every day, for li…
Mission
FamilyOS is building the world’s first AI Operating System for Families.
Our mission is to help every family make more confident decisions together, every day, for life.
Unlike traditional finance apps, calendars, note-taking tools, or AI chatbots, FamilyOS treats the household as the primary operating unit and provides AI-powered decision support across every important aspect of family life.
⸻
Vision
Families today manage their lives across dozens of disconnected apps:
* Banking
* Investments
* Insurance
* Calendars
* School emails
* Healthcare
* Notes
* Photos
* Travel
* Smart home devices
These tools store information, but they do not understand the family as a whole or help families make better decisions.
FamilyOS will become the trusted intelligence layer that connects every aspect of family life into one operating system.
Long-term, FamilyOS aims to become for families what ERP systems became for businesses.
Desired capability: The first product is FinanceOS + Family Today. FinanceOS acts as an AI Chief Financial Officer for households. Family Today is the family’s daily command center that answers one simple question: “What does our family need today?” The MVP focuses on: * Household financial management * Net worth * Cash flow * Investment portfolio * Retirement planning * Education funding * Family goals * Scenario simulation * AI financial recommendations * Daily family briefing
Success: FamilyOS is a Decision Operating System.
Every feature must help families make better decisions together.
If a feature does not improve family decision-making, it does not belong in FamilyOS.
Why Marrow thinks this
Your saved knowledge is strong on the AI-engineering half of this goal: multi-agent orchestration (Claude Cowork, multi-agent teams), agent decomposition (tools/skills/subagents), persistent LLM memory (Cognee), the prototype→production→optimization lifecycle, and AI-native first-principles product thinking. You also hold real, grounded personal-finance decision content: '持續買進' (systematic investing beats timing/selection), corporate-bond quality-over-yield analysis, and the '2倍法則'. The frontier gap is the bridge between these two clusters: you have agent architecture AND finance content, but almost nothing on how the *household* becomes the modeled unit — no notes on family power dynamics, joint decision-making, financial data aggregation, or how to prove a recommendation actually improved a family's decision. That measurement/validation gap is what blocks the success definition, not the engineering.
What you have explored (6)
Estimates from your saved notes — exposure, not mastery.
Multi-agent orchestration for specialized rolesfamiliar · 3 note(s)
· grounded
Family as the modeled decision unitNo saved note addresses households, joint decision-making, or family power dynamics. All agent and memory notes assume a single user or engineering team. This is the central modeling gap between your knowledge and the goal. · inferred
Decision-quality measurement / evaluation for recommendationsThe three-stage note mentions optimization/evaluation frameworks abstractly, but you have no note on measuring decision quality or outcome attribution — precisely what your success criterion requires. · inferred
Important connections (4)
Persistent memory layer for stateless LLMs→ prerequisite for →Family as the modeled decision unit
FinanceOS must remember a family's evolving finances, goals, and past decisions across sessions and members. Without a Cognee-style persistent, queryable memory keyed to the household, the 'family as the modeled unit' concept has no substrate — the LLM would reset each interaction and could never track a shared, multi-person financial state.
Multi-agent orchestration for specialized roles→ bridges to →系統性投資 vs 擇時選股 (systematic investing)
This is the missing bridge: your agent-orchestration knowledge is the delivery mechanism, and your finance content is the substance. A cash-flow agent, a portfolio agent, and a retirement-planning agent each need to embed concrete financial logic like systematic investing and credit analysis. Connecting the two clusters is exactly how the FinanceOS 'household CFO' becomes real rather than a generic chatbot.
Prototype → production → optimization lifecycle→ prerequisite for →Decision-quality measurement / evaluation for recommendations
Stage 3 optimization in the lifecycle depends on having an evaluation framework. For FinanceOS that framework must measure decision quality, not model accuracy — so defining the decision metric is a prerequisite for ever reaching the optimization stage the note describes.
AI agent autonomy spectrum→ applies to →Family as the modeled decision unit
The tools→skills→subagents spectrum determines how much financial autonomy you delegate on a family's behalf. In a household with money at stake and competing member interests, over-delegating to autonomous subagents raises the exact 'gradual power transfer' risk your existential-risk note flags. Matching autonomy level to family decision stakes is a design question unique to this goal.
Assumptions to challenge (1)
Decision-quality measurement / evaluation for recommendations⚡ in tension with ⚡AI-native first-principles product design
First-principles product thinking pushes you to build bold AI-native features fast, but your own success definition says nothing ships unless it *measurably* improves family decisions. Without a decision-quality metric, you cannot tell whether an AI recommendation was good or merely confident — a blind spot that could let FamilyOS confidently deliver harmful financial advice to families. This tension must be resolved before scaling.